This study examined a total of 110,792 individuals (46,465 men and 64,327 women aged 40 to 79) who participated in the Japan Collaborative Cohort Study for Evaluation of Cancer Risk Sponsored by the Ministry of Education, Science, Sports and Culture (JACC Study). During 1988–1990, individuals living in 45 communities across Japan voluntarily participated in this study and completed self-administered questionnaires including lifestyles and medical histories of previous cardiovascular disease and cancer at baseline. Details of the study procedure were described previously [17]. Informed consent was obtained prior to completing the questionnaire. We used participant's responses to a question regarding marital status: "What is your marital status (married, widowed, divorced or single)?" However, a total of 1,486 men and 2,015 women were excluded from the analysis because the question on marital status was omitted in two communities (4%). A total of 39,471 men and 54,591 women (aged 40 to 79) provided a valid response to this question, and the response rate to this question was 88% for both men and women. Of this group, we also excluded 1,690 men and 2,308 women from the analysis due to a positive history of stroke, coronary heart disease or cancer at baseline. Therefore, the responses of 37,781 men and 52,283 women were included in the analysis of this study.
Mortality surveillance
The subjects were followed from the date of the acceptance of the baseline survey through December 31, 1999. Since the information of the resident register is open to the general public under the resident registration law, investigators confirmed yearly residence status and survival with using residential registers, kept by a public health center in each of the study areas. Residency and death registration is required by Family Registration Law in Japan, and was believed to be complete across Japan. Death certificate diagnoses were provided by the Ministry of Health and Labor under permission from Welfare after Ministry of Internal Affairs and Communications granted permission. The underlying causes of deaths were defined according to the International Classification of Diseases, 9th Revision from 1988 to 1994, and 10th Revision from 1995 to 1999 for the National Vital Statistics. Therefore, all deaths that occurred in the cohort were confirmed by death certificates from a public health center, except for subjects who died after they moved from their original community, in which case the subject was treated as a censored case. The average follow-up period for the participants was 9.9 years. The Ethical Committees of the Nagoya University School of Medicine and the University of Tsukuba approved the present study.
Statistical analysis
Statistical analyses were based on gender-stratified mortality during the follow-up period from 1989 to 1999. Each participant contributed person-time from the date of completed baseline questionnaire until the time of death or relocation, or December 31, 1999. Age-adjusted means and proportions of selected mortality risk factors were presented among the categories of marital status; the statistical testing for differences among marital status was conducted using analysis of covariance. Cox proportional hazards modeling was used to determine whether the marital status was significantly associated with stroke (ICD-9 codes 430 to 438, ICD-10 codes I60 to I69), coronary heart disease (410 to 414, I20 to I25), cardiovascular disease (390 to 459, I01 to I99), cancer (140 to 208, C00 to C97), respiratory disease (460 to 519, J00 to J99), external causes (800 to 999, S00 to T98) and all causes. We compared the mortality rates for people who were widowed, divorced, and single to those who were married (the reference group). The age-adjusted and multivariate-adjusted relative risks with 95% confidence intervals were calculated after adjustment for age and potential confounders. The confounding variables were obtained from the baseline questionnaire. Variables that were associated with both mortality and the categories of marital status were included in the multivariate analysis. The assumption of proportional hazards for all the selected confounders was tested by using both the time-dependent covariate method and the linear correlation test. We found no violation of the proportionality assumption. These confounding variables were age (in years), body mass index (gender-stratified quintiles), smoking status (never, ex-smoker, and current smokers of 1 to 19 and ≥20 cigarettes per day), alcohol intake (never, ex-drinker, and current drinkers of ethanol at 1 to 22, 23 to 45, 46 to 68, and ≥69 g per day), hours of walking (<1, 1 to 2, 3 to 4, and ≥5 hours per week), education (<13, 13 to 15, 16 to 18, ≥19 years), employment status (employed versus unemployed), and interest level in health screening (no, low, moderate, or high), having children (no or yes), history of hypertension (no versus yes) and history of diabetes (no versus yes). Psychological variables other than perceived mental stress were included as potential confounders in the additional analyses. These included hopelessness (definitely yes, yes, may be yes, or no), self-estimation of quick response (yes, may be yes, or no), sense of hurry (definitely yes, yes, may be yes, or no), anger (yes, may be yes, or no), joyfulness (definitely yes, yes, may be yes, or no), sense of being trusted (definitely yes, yes, may be yes, or no), and fulfillment (definitely yes, yes, may be yes, or no). Further stratification by age (40 to 64 years and ≥65 years) was also obtained in the analysis to assess effect modification. The presence of a statistical interaction was tested by using cross-product terms of marital status and stratifying variables. In addition, to reduce the potential effect of as-yet-undiagnosed diseases at baseline that could have affected marital status, we examined the marital status and mortality associations after exclusion of deaths during the first two years of follow-up.